Models & Labs
OpenAI Develops Custom Jalapeño Chip with Broadcom
TechCrunch AIhigh confidence
Why it matters
- →Custom chips can provide performance optimizations tailored to specific AI workloads.
- →Reducing dependency on Nvidia could lead to more competitive pricing and innovation in AI hardware.
- →OpenAI's move reflects a broader industry trend towards custom silicon for strategic control.
OpenAI has announced its plans to develop a custom inference chip called Jalapeño in collaboration with Broadcom. This move is part of a broader trend among tech giants to reduce dependency on Nvidia's GPUs by creating hardware tailored to specific needs. The Jalapeño chip is expected to provide OpenAI with more control and potentially significant performance improvements. This development places OpenAI alongside companies like Google and Apple, who are also pursuing custom silicon solutions.
Read originalMore from TechCrunch AI
© TechCrunch AIMarket & Regulationbusiness
Apple Vision Pro Exec Joins OpenAI Hardware Team
Paul Meade, a pivotal player in Apple's Vision Pro headset development, is transitioning to OpenAI's hardware team. This move comes as Apple undergoes leadership changes with John Ternus set to become CEO, leading to a reshuffle in the hardware engineering division. Meade's shift highlights the increasing allure of AI-driven companies for top tech talent. OpenAI is collaborating with former Apple design chief Jony Ive on a new AI device, aiming to create a more tranquil user experience compared to current smartphones. This development marks a significant step in the merging of AI and hardware innovation, as OpenAI seeks to redefine consumer technology.
TechCrunch AI
© TechCrunch AIModels & Labsmodels
Asian AI Startups Launch Models Amid Anthropic Ban
In a strategic move, Asian AI startups are stepping into the spotlight as the U.S. export ban on Anthropic's Mythos and Fable models continues. Chinese cybersecurity firm 360 has introduced Tulongfeng, an AI tool aimed at software vulnerability detection, while Tokyo-based Sakana AI has launched Fugu, a model designed for agent orchestration and optimized for Japanese language and culture. These launches highlight a growing trend of regional AI development, offering alternatives to U.S. models and addressing local needs. As the export ban persists, these startups are seizing the opportunity to fill the void left by restricted access to U.S. AI technologies.
TechCrunch AI
© TechCrunch AIMarket & Regulationmodels
Trump Admin Allows Anthropic Mythos 5 Use
In a significant policy shift, the Trump administration has permitted Anthropic to redeploy its Mythos 5 model to over 100 U.S. companies and agencies. This decision comes after a ban that restricted access due to security concerns. The move allows non-American employees at these organizations, as well as Anthropic's own non-American staff, to use the model. While the directive does not address the Fable 5 model, it marks a step towards broader access to Anthropic's cybersecurity tools. This change could enhance the cybersecurity capabilities of critical U.S. infrastructure.
TechCrunch AI
More in Models & Labs
Models & Labsmodels
llama.cpp b9817 release enhances OpenVINO support
The latest b9817 release of llama.cpp brings significant updates to its OpenVINO backend, including an upgrade to OV 2026.2.1 and the introduction of self-contained release packages. These changes streamline the deployment process and improve operator handling, making it easier for developers to integrate and utilize OpenVINO in their projects. Additionally, the update removes hardcoded compute operation types, enhancing flexibility and adaptability. This release marks a step forward in making llama.cpp a more versatile and developer-friendly platform, particularly for those leveraging OpenVINO's capabilities.
llama.cpp Releases
Models & Labsmodels
llama.cpp b9820 Release Enhances CUDA Performance
The b9820 release of llama.cpp brings notable improvements to CUDA performance by cutting down on unnecessary synchronizations, which can streamline token processing. This update introduces asynchronous copy capabilities between CPU and CUDA, facilitating smoother data transfers and potentially speeding up computations. Backend detection has been refined to avoid linking conflicts, and synchronization adjustments have been made more general, allowing other backends like Vulkan to benefit. These enhancements aim to optimize performance across different hardware setups, making llama.cpp a more adaptable tool for developers working with diverse configurations.
llama.cpp Releases
Models & Labsmodels
llama.cpp b9826 Release Expands Platform Support
The b9826 release of llama.cpp continues to enhance its reach by supporting a wider array of systems, though it doesn't bring new model architectures. With ROCm 7.2 now available for Ubuntu x64, AMD GPU users gain a viable alternative to NVIDIA's CUDA, broadening their options for AI inference. The update also includes builds for macOS, Linux, Windows, and openEuler, ensuring developers can utilize llama.cpp regardless of their operating environment. While the release doesn't introduce groundbreaking features, it reinforces llama.cpp's utility as a flexible tool for AI developers working across different hardware and software configurations.
llama.cpp Releases